Re: [PATCH] sched/topology: Correct "sched_domains_curr_level" in topology_span_sane()
From: K Prateek Nayak
Date: Thu Jun 26 2025 - 00:24:44 EST
Hello Valentin,
Thank you for taking a look at the patch.
On 6/25/2025 9:02 PM, Valentin Schneider wrote:
Hey,
First of all, thanks for looking into this!
On 24/06/25 04:12, K Prateek Nayak wrote:
The updated topology_span_sane() algorithm in commit ce29a7da84cd
("sched/topology: Refinement to topology_span_sane speedup") works on
the "sched_domain_topology_level" hierarchy to detect overlap in
!SDTL_OVERLAP domains using the tl->mask() as opposed to the
sched_domain hierarchy (and the sched_domain_span()) in the previous
approach.
The previous approach also used tl->mask() directly, but it happened
to be called *before* the build_sched_domain() loop, so the NODE iteration
happened with sched_domain_curr_level at its default static value of
0... For the first SD build that is, I assume that was then broken for any
subsequent rebuild.
Oh right! I read it wrong. My bad. topology_span_sane() being called
before build_sched_domain() always worked on "sched_domains_curr_level"
of the previous domain and could potentially cause an issue during
rebuild if we start from the NODE domain.
For NODE domain, the cpumask retunred by tl->mask() depends on the
"sched_domains_curr_level". Unless the "sched_domains_curr_level" is
reset during topology_span_sane(), it reflects the "numa_level" of the
last sched_domain created which is incorrect.
Reset the "sched_domains_curr_level" to the "tl->numa_level" during
topology_span_sane(). Although setting "topology_span_sane" to 0 in
topology_span_sane() should be enough since all domains with
numa_level > 0 currently set SDTL_OVERLAP flag, setting it to
"tl->numa_level" makes it more explicit and future proof against changes
in the same area.
Cc: Steve Wahl <steve.wahl@xxxxxxx>
Reported-by: Leon Romanovsky <leon@xxxxxxxxxx>
Closes: https://lore.kernel.org/lkml/20250610110701.GA256154@unreal/
Fixes: ce29a7da84cd ("sched/topology: Refinement to topology_span_sane speedup")
Per the above, this could probably be:
Fixes: ccf74128d66c ("sched/topology: Assert non-NUMA topology masks don't (partially) overlap")
I'll update this in the next version.
Signed-off-by: K Prateek Nayak <kprateek.nayak@xxxxxxx>
---
This issue can be reproduced on a setup with the following NUMA topology
shared by Leon:
$ sudo numactl -H
available: 5 nodes (0-4)
node 0 cpus: 0 1
node 0 size: 2927 MB
node 0 free: 1603 MB
node 1 cpus: 2 3
node 1 size: 3023 MB
node 1 free: 3008 MB
node 2 cpus: 4 5
node 2 size: 3023 MB
node 2 free: 3007 MB
node 3 cpus: 6 7
node 3 size: 3023 MB
node 3 free: 3002 MB
node 4 cpus: 8 9
node 4 size: 3022 MB
node 4 free: 2718 MB
node distances:
node 0 1 2 3 4
0: 10 39 38 37 36
1: 39 10 38 37 36
2: 38 38 10 37 36
3: 37 37 37 10 36
4: 36 36 36 36 10
This topology can be emulated using QEMU with the following cmdline used
in my testing:
sudo ~/dev/qemu/build/qemu-system-x86_64 -enable-kvm \
-cpu host \
-m 20G -smp cpus=10,sockets=10 -machine q35 \
-object memory-backend-ram,size=4G,id=m0 \
-object memory-backend-ram,size=4G,id=m1 \
-object memory-backend-ram,size=4G,id=m2 \
-object memory-backend-ram,size=4G,id=m3 \
-object memory-backend-ram,size=4G,id=m4 \
-numa node,cpus=0-1,memdev=m0,nodeid=0 \
-numa node,cpus=2-3,memdev=m1,nodeid=1 \
-numa node,cpus=4-5,memdev=m2,nodeid=2 \
-numa node,cpus=6-7,memdev=m3,nodeid=3 \
-numa node,cpus=8-9,memdev=m4,nodeid=4 \
-numa dist,src=0,dst=1,val=39 \
-numa dist,src=0,dst=2,val=38 \
-numa dist,src=0,dst=3,val=37 \
-numa dist,src=0,dst=4,val=36 \
-numa dist,src=1,dst=0,val=39 \
-numa dist,src=1,dst=2,val=38 \
-numa dist,src=1,dst=3,val=37 \
-numa dist,src=1,dst=4,val=36 \
-numa dist,src=2,dst=0,val=38 \
-numa dist,src=2,dst=1,val=38 \
-numa dist,src=2,dst=3,val=37 \
-numa dist,src=2,dst=4,val=36 \
-numa dist,src=3,dst=0,val=37 \
-numa dist,src=3,dst=1,val=37 \
-numa dist,src=3,dst=2,val=37 \
-numa dist,src=3,dst=4,val=36 \
-numa dist,src=4,dst=0,val=36 \
-numa dist,src=4,dst=1,val=36 \
-numa dist,src=4,dst=2,val=36 \
-numa dist,src=4,dst=3,val=36 \
...
It's a bit of a mouthful but I would keep that in the changelog itself
given that it's part of reproducing the bug.
Sure thing!
Changes are based on tip:sched/urgent at commit 914873bc7df9 ("Merge tag
'x86-build-2025-05-25' of
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
---
kernel/sched/topology.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index a2a38e1b6f18..1d634862c8df 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -2426,6 +2426,15 @@ static bool topology_span_sane(const struct cpumask *cpu_map)
cpumask_clear(covered);
cpumask_clear(id_seen);
+#ifdef CONFIG_NUMA
+ /*
+ * Reuse the sched_domains_curr_level hack since
+ * tl->mask() below can resolve to sd_numa_mask()
+ * for the NODE domain.
+ */
+ sched_domains_curr_level = tl->numa_level;
+#endif
+
Urgh... Given this is now invoked after the build_sched_domain() loop, what
if we directly used the sched_domain_span(), instead, i.e. use
sched_domain_mask(per_cpu_ptr(tl->data->sd, cpu))
instead of
tl->mask(id)
which means no indrect use of sched_domains_curr_level. Note that I'm
currently running out of brain juice so this might be a really stupid idea :-)
Let me go try that! It should also help detect overlap in case
build_sched_domain() has fixed up domain spans when the child's span is
not a subset of the parent's span.
/*
* Non-NUMA levels cannot partially overlap - they must be either
* completely equal or completely disjoint. Otherwise we can end up
base-commit: 914873bc7df913db988284876c16257e6ab772c6
--
2.34.1
--
Thanks and Regards,
Prateek